【作者】 张凯；

【导师】 赵宇红；

【作者基本信息】 内蒙古科技大学 ， 计算机技术（专业学位）， 2021， 硕士

【摘要】 社区发现是复杂网络分析中的重要研究内容,广泛应用于推荐系统、广告投放和舆论监测等方面。但是现在大多数社区发现方法的研究,都是基于同构信息网络(Homogeneous information network)展开,即将网络中所有的节点都看做是一种类型。但是实际生活中的大多数网络类型都是异构信息网络(Heterogeneous information network),即网络中的节点和边都属于同一种类型。最近几年来,异构信息网络的研究受到了学术界的广泛关注,为了更好地研究异构信息网络,本文针对异构信息网络中的社区发现研究算法进行了深入研究,具体研究内容包括:1、相似性度量:为了更准确地进行节点之间的相似性度量,首先通过信息散度得到相邻两节点之间的相似性,并结合注意力机制得到对应节点下元路径的权重,将二者进行加权融合,提出一种全新的节点相似性度量方法。2、结合改进的K-means聚类算法进行社区发现:传统的K-means聚类算法获取聚类中心的办法是采用的随机选取的方法,聚类中心选取的准确性很重要,不合适的聚类中心会导致社区划分结果不准确。基于此,本文结合新的相似性度量算法,提出了一种全新的聚类中心选取的方法,不再随机地选取聚类中心,从而使得社区发现的划分结果更准确。本文通过在三个异构信息网络数据集DBLP,ACM和IMDB下进行实验,并且与传统的PathSim,AvgSim等算法进行对比,以此来验证所提出的相似性度量算法是否准确。然后将改进的算法与K-means聚类算法做结合,仿真实验结果表明,基于改进后的K-means聚类算法无论是在NMI还是模块度上均有明显的提升,仿真实验结果也验证了基于改进后的K-means算法可以有效地处理异构信息网络,方法可行、有效。更多还原

【Abstract】 Community discovery is an important research content in complex network analysis,which is widely used in recommendation system,advertising and public opinion monitoring.But now most of the research on community discovery methods are based on the homogeneous information network,that is,all the nodes in the network are regarded as a type.But most of the network types in real life are heterogeneous information network,that is,the nodes and edges in the network belong to the same type.In recent years,the research of heterogeneous information network has been widely concerned by the academic community.In order to better study the heterogeneous information network,this paper makes an in-depth study on the community discovery algorithm in heterogeneous information network.The specific research contents include:1.Similarity measurement: in order to measure the similarity between points more accurately,firstly,the similarity between two adjacent nodes is obtained by information divergence,and the weight of the meta path under the corresponding node is obtained by combining the attention mechanism.Then the two are weighted and fused,and a new node similarity measurement method is proposed.2.Combined with the improved k-means clustering algorithm for community discovery: the traditional K-means clustering algorithm to obtain the cluster center is the method of random selection,the accuracy of cluster center selection is very important,inappropriate cluster center will lead to community division results error.Based on this,combined with the new similarity measurement algorithm,this paper proposes a new method to select the cluster center,no longer randomly select the cluster center,so as to make the partition result of community discovery more accurate.更多还原